Infinitus Systems Interview Guide (2026): Voice Agents That Call Payers

Infinitus Systems was founded in 2019 in San Francisco by Ankit Jain, a former Google engineering director and founding partner at Gradient Ventures, and Shyam Rajagopalan, now CTO, after Google’s Duplex demo made them ask whether an AI could sit on hold with a payer instead of a human. The product is a voice agent (internally named Eva) that makes outbound calls to payers, pharmacies, and providers, plus a copilot (FastTrack), an analytics layer (Lens), a no-code agent builder (Studio, launched April 2026), and an AI-first patient-support hub announced in July 2026. The company closed a $51.5M Series C led by Andreessen Horowitz in October 2024, bringing total funding to $102.9M alongside GV, Kleiner Perkins, Coatue, and Memorial Hermann; its own site says the company is valued at over $600 million, and third-party trackers put the Series C post-money just under that. Trackers put headcount between 110 and 215. Engineering Lead Gurudutta Ramanathaiah came from Twilio, Sonder, and Sigma Computing.

Process

Infinitus does not publish its loop, so what follows is stitched from Glassdoor reports through 2025 and 2026 and from the shape of its job postings. Reports agree on the front half: a recruiter call, then a one-hour technical screen where you write pseudocode for a use case and are asked how you would scale it to a system, then a second hour split into roughly 20 minutes of behavioral questions and 40 minutes of coding at LeetCode-medium level. One candidate reported a take-home instead of the second screen, followed by three onsite interviews in a half day. The fuller report describes a five-round onsite; for a frontend-leaning role that meant an hour of frontend system design, an hour of React coding, an hour of backend design centered on API design and database scaling, a 30-minute product-infrastructure conversation that was mostly behavioral, and a 30-minute culture-fit round. After a positive onsite there is a 15-minute chat with Ankit Jain. Glassdoor difficulty averages 2.7 out of 5, and about half of reviewers rate the experience positively; the complaints are about scheduling gaps, not content. Timelines run from one week to two months.

Stage	Who	What it tests
Recruiter screen	Recruiter	Background, hybrid fit (Mon, Tue, Thu in the SF office), comp expectations
Technical screen one, 60 min	Engineer	Pseudocode for a realistic use case, then how you would scale it
Technical screen two, 60 min	Engineer	About 20 min behavioral, 40 min data structures and algorithms at LeetCode medium
Onsite, 3 to 5 rounds	Engineers and product	Frontend or backend design, live coding in your stack (React for frontend roles), API and database scaling, product infrastructure, culture fit
Founder chat, 15 min	Ankit Jain, CEO	Your questions about the company; a light final check on agency and urgency

The stack you will be coding against is Go on Google Cloud (Firestore, PostgreSQL, BigQuery, Pub/Sub) with a React, TypeScript, GraphQL, and WebSockets frontend, per the mid-career Full Stack posting that came down in January 2026; the Forward Deployed Engineer posting wants Python, production LLM experience, and MCP servers. On the difficulty index this is a well-run Series C, not a research lab: fewer rounds than Abridge or Hippocratic, more weight on shipping. Jain has written that he hires for agency and urgency, and employee reviews say there are no manager, director, or VP titles, so expect to be asked what you did when nobody assigned you the work.

What they actually ask

• Design the system that lets an agent place ten thousand concurrent outbound calls to payer IVRs, hold for forty minutes, and hand a structured benefit-verification result back to a customer’s CRM. Where does state live when a call drops at minute thirty-eight?
• A call to a payer can run over three hours and three hundred conversational turns. How would you detect, in real time, that the conversation has broken down (wrong department, agent confusion, silence) using both the audio and the transcript, and what action does the agent take?
• Given a payer, plan, treatment, and patient demographics, write the query that decides whether we already have a verified answer and can skip the call, or need to dial. Now explain why Infinitus moved that logic out of JSON blobs in Postgres into a graph, and what you would have chosen.
• The agent must never state a coverage fact it did not hear on the call. Describe a guardrail design where the LLM chooses from a fixed set of allowed actions instead of free-generating, and show how you would test that the model cannot escape that set.
• You have 300 million human-reviewed utterances and a fine-tuned in-house model. How do you decide which calls go to human QA, how do you measure whether data accuracy improved 10 percent over human callers, and what is your confidence interval?
• Design the API a pharma hub uses to submit a batch of re-verification requests each January, when volume spikes 144 percent. Cover idempotency, rate limiting against payer phone lines, and how results get back to the customer.
• Every recording and transcript contains PHI. Explain how you would partition storage and access so a customer-success engineer can debug a failed call without seeing member IDs, and what audit trail you leave behind.
• Tell me about a time you shipped something nobody asked you to build. What did it cost you, and did it turn out to matter?

Levels and comp (2026)

Infinitus does not use a public leveling ladder, and employee reviews say the company avoids management titles altogether, so “level” in an offer mostly means the scope of the role and the size of the grant. The one posted band we could confirm is on the AI Full Stack Engineer listing (San Francisco, hybrid), which shows $130,800 to $186,200 in base salary. On levels.fyi the median reported software engineer package is around $189K total in the Bay Area with a full-stack median near $190K, and the top reported package sits around $427K, so senior offers can clear the posted band once equity is counted. Glassdoor’s 85-salary sample and public H-1B filings cluster engineer base pay between $150K and $200K. Check levels.fyi and Blind for fresh data before you negotiate.

Equity vests over four years. The number that matters is the preference stack: about $103M of preferred sits ahead of your options, and the last priced round valued the company in the high hundreds of millions, so your grant pays only on an exit above that. Ask for the strike price, the latest 409A, the fully diluted share count, and whether early exercise is allowed. No raise has been announced since October 2024; ask the recruiter whether that means profitability or a Series D in progress, because the answer changes what the grant is worth.

Prep priorities

Learn the payer-call domain first, because every design question is really about it: what a benefit verification captures (deductible, out-of-pocket max, prior-auth requirement, site-of-care rules, specialty pharmacy routing), what a prior-auth follow-up call sounds like, and why re-verification piles up every January. Then read the two technical pieces the company has published: the “Infinitus AI machine” blog post on the 100-plus-model stack and the multimodal dialogue-breakdown detector, and the Neo4j customer story where Ramanathaiah explains the payer-plan-treatment graph. They hand you the vocabulary the interviewers use. On the technical side, get Go or Python fluent enough to write a queue consumer and a retry loop without thinking, review system design patterns for long-lived stateful workflows (durable state machines, idempotent handlers, backpressure against a scarce resource, here a payer’s phone line), and do a week of LeetCode mediums on strings, hashing, and intervals.

Ask the recruiter early which loop variant you are getting (two screens or a take-home), whether the onsite is three rounds or five, and who besides Jain you meet at the end. Ask how the hybrid schedule is enforced; postings name three set office days and reviews suggest that is real. Frontend candidates should confirm React is the live-coding language before spending prep on backend design. Forward Deployed candidates should expect customer judgment and LLM tooling to outweigh algorithms, and come ready to walk through a deployment they owned end to end.

Frequently Asked Questions

How hard is the Infinitus Systems coding interview?

Candidates on Glassdoor rate it 2.7 out of 5, which is average for a Series C startup. The coding portion is 40 minutes at LeetCode-medium level, and the harder rounds are the design ones, where interviewers push on scaling long-running calls and on database choices.

Does Infinitus give a take-home assignment?

Some candidates report one, in place of the second technical screen, followed by a half-day onsite of three interviews. Others describe two live screens and a five-round onsite with no take-home. Ask the recruiter which format applies to your role before you start.

Is Infinitus remote or hybrid?

Engineering roles are hybrid in San Francisco, with Monday, Tuesday, and Thursday listed as in-office days on recent postings. Go-to-market, implementation, and some Forward Deployed Engineer roles are remote within the US.

What does Infinitus pay software engineers?

The only posted band is $130,800 to $186,200 base for an AI Full Stack Engineer in San Francisco. Levels.fyi shows a median software engineer package near $189K and a top reported package around $427K. Equity vests over four years, and the company last priced its shares in the October 2024 Series C, so ask for the strike price and 409A before you compare against a public-company offer.